It's just an idea of course. I found out this open source codec that works *very well* for voices (did some tests and reduces text lines much better than ogg). http://www.speex.org/
Problem is that needs to be added to renpy, but I found that there are already some python bindings: http://www.freenet.org.nz/python/pySpeex/
of course it's completely up to PyTom if he wants to add it. I don't know how many people here use voices in their games, but since they're quite common in VNs, I think might be a very nice feature to add!
I actually removed it a while back, since nobody used it and it was moderately large. Also, the most speex files were low quality... have you found settings that make them sound better than vorbis at the same bitrate?
Supporting creators since 2004 (When was the last time you backed up your game?)
Hm true, I looked better at the website. Seems like the "best" setting they suggest is at 32khz. I made some tests but really even for voiceovers (especially female voices) you need to use 44khz otherwise sounds bad on high pitches like when screaming/yelling (I can notice it immediately).
Well there's no solution for Bionic Heart then, 25mb of voiceovers but at least managed to do the download-file from within Renpy work!
Yes, I use oggdropXpd (that fun thing where you drag&drop the files). Actually the fact is that there's lot of text in the game. Only the demo, is 40:28 of voices with over 700+ files (and isn't yet 100% voiced, but like 90%)!!
I suspect that a reasonable amount of size could be saved by proper conversion of the voice files. For example, voice is generally coming from the center of the screen, so it might not be inappropriate to convert it to mono rather then stereo. I also find that vorbis often retains a reasonable amount of quality even at relatively low settings.
Supporting creators since 2004 (When was the last time you backed up your game?)
As I understand it speex is usefull for listening to talks, but isn't meant to preserve quality of sound (cuts highs and lows I believe). So if you want to character actor's voices clear you might want to stick with vorbis.
"The universe is non-simultaneously apprehended"
— Buckminster Fuller
PyTom wrote:I suspect that a reasonable amount of size could be saved by proper conversion of the voice files. For example, voice is generally coming from the center of the screen, so it might not be inappropriate to convert it to mono rather then stereo. I also find that vorbis often retains a reasonable amount of quality even at relatively low settings.
Yes of course I've converted all sounds to mono
I tried compressing with oggenc at quality setting 2. If I go below, like 1 or 0, especially on female voices, the difference is very noticeable (I can notice it with just normal speakers). I might try with male voices though, could work better.
Voices are and will always be huge. Voice codecs like Speex are optimized for being understandable, not high fidelity - using them in a game will kill immersion.
but at least managed to do the download-file from within Renpy work!
